基于可解释集成学习的异常驾驶行为风险识别方法

doi:10.16097/j.cnki.1009-6744.2025.02.017

摘要/Abstract

摘要： 为实现对异常驾驶行为风险的准确识别，并克服现有模型可解释性差的局限，本文通过自然驾驶试验获取车辆运动数据，针对超速、急变速、急转弯、跟车过近、危险变道这5种异常驾驶行为，采用阈值法量化其风险系数，结合CRITIC(CriteriaImportance Through Inter-criteria Correlation)权重法及分位值法划分异常驾驶行为风险等级。构建基于Stacking的集成学习识别模型，由不同学习器组合训练结果，选择综合性能最佳的学习器GBDT(Gradient Boosting Decision Tree)、AdaBoost、XGBoost作为初级学习器组合，采用Logistic回归作为次级学习器。在此基础上，利用SHAP(ShapleyAdditiveexPlanation)算法分析了特征变量对最优Stacking模型识别结果的影响。研究表明，最优Stacking模型识别准确率达92.68%，对于异常驾驶行为风险实现较高精度的识别。车速和换道碰撞时间是对模型识别结果影响较大的特征，车速超过95 km·h-1和换道碰撞时间小于2.8s均会增加行为风险，且车速超过150km·h-1时风险等级更高。本文为异常驾驶行为风险的识别与解释提供了一个可行的框架，有望为提升交通安全水平提供技术支持。

关键词: 交通工程, 风险识别, Stacking集成学习, 异常驾驶行为, 自然驾驶试验, SHAP算法

Abstract: In order to accurately identify the risk of abnormal driving behavior and overcome the limitations of poor interpretability of existing models, this study collected vehicle motion data through natural driving test. Five types of abnormal driving behaviors were investigated: speeding, rapid shifting, sharp turning, short distance following, and dangerous lane changing. The risk coefficients of these behaviors were quantified using a threshold method, and the risk levels were classified using the CRITIC (Criteria Importance Through Inter-criteria Correlation) weight method and quantile method. A Stacking based ensemble learning identification model was constructed to identify the abnormal driving behaviors. The model combined training results from different learners. GBDT (Gradient Boosting Decision Tree), AdaBoost, and XGBoost that have the best comprehensive performance, selected as the primary learner combination, and logistic regression was used as the secondary learner. The SHAP (Shapley Additive exPlanation) algorithm was then used to analyze the influence of feature variables on the recognition results of the optimal Stacking model. Results indicate that the optimal Stacking model has an identification accuracy of 92.68%, achieving high precision in identifying abnormal driving behavior risks. Vehicle speed and time-to-collision of lane changing were identified as key features significantly impacting model recognition. Specifically, vehicle speed exceeding 95 km·h^-1 and time-to-collision of lane changing less than 2.8 s both increase behavioral risk, and the risk level is higher when the vehicle speed exceeds 150 km·h^-1. This study provides a feasible framework for identifying and interpreting the risks of abnormal driving behavior, which is expected to provide technical support for improving traffic safety levels.

Key words: traffic engineering, risk identification, Stacking ensemble learning, abnormal driving behavior, natural driving test, SHAPalgorithm

中图分类号:

U491

邓院昌, 蒋昀轩, 陶胜芹. 基于可解释集成学习的异常驾驶行为风险识别方法[J]. 交通运输系统工程与信息, 2025, 25(2): 180-189.

DENG Yuanchang, JIANG Yunxuan, TAO Shengqin. Risk Identification Method for Abnormal Driving Behavior Based on Interpretable Ensemble Learning Model[J]. Journal of Transportation Systems Engineering and Information Technology, 2025, 25(2): 180-189.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: http://www.tseit.org.cn/CN/10.16097/j.cnki.1009-6744.2025.02.017

http://www.tseit.org.cn/CN/Y2025/V25/I2/180

参考文献

[1]MA Y, TANG K, CHEN S, et al. On-line aggressive driving identification based on in-vehicle kinematic parameters under naturalistic driving conditions[J]. Transportation Research Part C: Emerging Technologies, 2020, 114: 554-571.

[2] 程国柱,李天仪,汪国鹏.基于冰雪路面危险驾驶行为谱的行车风险识别方法[J].交通运输系统工程与信息, 2024, 24(4): 127-138. [CHENG G Z, LI T Y, WANG G P, Driving risk identification method based on dangerous driving behavior spectrum on ice and snow pavement[J]. Journal of Transportation Systems Engineering and Information Technology, 2024, 24(4): 127-138.]

[3] 吴建清,张子毅,王钰博,等.考虑多模态数据的重载货车危险驾驶行为识别方法[J].交通运输系统工程与信息, 2024, 24(2): 63-75. [WU J Q, ZHANG Z Y, WANG Y B, et al. Method for identifying dangerous driving behaviors in heavy-duty trucks based on multi modal data[J]. Journal of Transportation Systems Engineering and Information Technology, 2024, 24(2): 63-75.]

[4]陆键,王可,蒋愚明.基于车辆行驶轨迹的道路不良驾驶行为实时辨识方法[J].交通运输工程学报,2020,20 (6): 227-235. [LU J, WANG K, JIANG Y M. Real-time identification method of abnormal road driving behavior based on vehicle driving trajectory[J]. Journal of Traffic and Transportation Engineering, 2020, 20(6): 227-235.]

[5]王雪松,徐晓妍.基于自然驾驶数据的危险事件识别方法[J]. 同济大学学报(自然科学版),2020,48(1): 51 59. [WANG X S, XU X Y. Detection of safety-critical events based on naturalistic driving data[J]. Journal of Tongji University (Natural Science), 2020, 48(1): 51-59.]

[6] SHI X, WONG Y D, LI M Z F, et al. A feature learning approach based on XGBoost for driving assessment and risk prediction[J]. Accident Analysis & Prevention, 2019, 129: 170-179.

[7]单永航,张希,胡川,等.基于集成学习的交通事故严重程度预测研究与应用[J].计算机工程,2024,50(2): 33-42. [SHAN Y H, ZHANG X, HU C, et al. Traffic accident severity prediction research and application based on ensemble learning[J]. Computer Engineering, 2024, 50(2): 33-42.]

[8] DU X J, ZHAO W. Risky lane-changing behavior recognition based on stacking ensemble learning on snowy and icy surfaces[J]. Scientific Reports, 2024, 14 (1): 19257.

[9] GUO M, ZHAO X, YAO Y, et al. A study of freeway crash risk prediction and interpretation based on risky driving behavior and traffic flow data[J]. Accident Analysis & Prevention, 2021, 160: 106328.

[10] MA Y, ZHANG J, LU J, et al. Prediction and analysis of likelihood of freeway crash occurrence considering risky driving behavior[J]. Accident Analysis & Prevention. 2023, 192: 107244.

[11] 刘通, 付锐, 张士伟,等.车辆典型危险行驶状态识别与检测研究进展[J].中国安全科学学报,2017,27(10): 32-37. [LIU T, FU R, ZHANG S W, et al. Progress in research on identification and detection of vehicle typical hazardous driving states[J]. China Safety Science Journal, 2017, 27(10): 32-37.]

[12] 王可, 陆键,蒋愚明.基于车辆行驶轨迹的道路不良驾驶行为谱构建与特征值计算方法[J].交通运输工程学报, 2020, 20(6): 236-249. [WANG K, LU J, JIANG Y M. Abnormal road driving behavior spectrum establishment and characteristic value calculation method based on vehicle driving trajectory[J]. Journal of Traffic and Transportation Engineering, 2020, 20(6): 236-249.]

[13] 薛清文,蒋愚明,陆键.基于轨迹数据的危险驾驶行为识别方法[J]. 中国公路学报,2020,33(6): 84-94. [XUE Q W, JIANG Y M, LU J. Risky driving behavior recognition based on trajectory data[J]. China Journal of Highway and Transport, 2020, 33(6): 84-94.]

[14] 陈必俊.基于交通轨迹大数据的驾驶行为安全性评价方法[D]. 福州: 福建工程学院,2021. [CHENBJ.Safety evaluation for driving behavior based on massive traffic trajectory data[D]. Fuzhou: Fujian University of Technology, 2021.]

[15] 交通运输部公路局.公路安全生命防护工程实施技术指南[M]. 北京: 人民交通出版社, 2015. [Highway Administration Bureau of Ministry of Transport. Guide for implementation of improve highway safety to cherish the life project[M]. Beijing: China Communication Press, 2015.]

[16] DOUZAS G, BACAO F, LSAT F. Improving imbalanced learning through a heuristic oversampling method based on k-means and SMOTE[J]. Information Sciences, 2018, 465: 1-2

[17] 周星, 丁立新,万润泽,等.分类器集成算法研究[J].武汉大学学报(理学版),2015,61(6): 503-508. [ZHOU X, DING L X, WAN R Z, et al. Research on classifier ensemble algorithms[J]. Journal of Wuhan University (Natural Science Edition), 2015, 61(6): 503-508.]

[18] LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions[J]. Arxiv Preprint Arxiv: 1705.07874, 2017.

[19] 张韡, 吴晓多,白骞,等.基于多项Logit模型的山区高速公路事故致因分析[J].科学技术与工程,2024,24 (5): 2111-2117. [ZHANG W, WU X D, BAI Q, et al. Analysis of direct causes of highway accidents in mountainous areas based on multinomial Logit model[J]. Science Technology and Engineering, 2024, 24(5): 2111 2117.]

[20] 李长城, 刘小明,荣建,等.不同能见度条件下高速公路车辆速度特性研究[J].交通运输系统工程与信息, 2014, 14(6): 213-218. [LI C C, LIU X M, RONG J, et al. Characteristics of vehicle speed for expressway under different visibility condition[J]. Journal of Transportation Systems Engineering and Information Technology, 2014, 14(6): 213-218.]

[21] 南彦洲, 柯辉,朱才华,等.面向驾驶员心电数据的交叉口驾驶风险评估方法[J].中国安全科学学报,2023, 33(2): 185-193. [NAN Y Z, KE H, ZHU C H, et al. A driving risk assessment method at intersection using driver's ECG data[J]. China Safety Science Journal, 2023, 33(2): 185-193.]